Related Experiment Video
Updated: Jan 14, 2026
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
Current concerns and future directions of large language model ChatGPT in medicine: a machine-learning-driven
Song-Bin Guo1,2,3, Deng-Yao Liu4,5, Xiao-Jie Fang1,3
1Department of Medical Oncology, Sun Yat-sen University Cancer Center, Guangzhou, P. R. China.
Background:
In its infancy, large language model (LLM) Chat Generative Pre-trained Transformer (ChatGPT) has delivered significant transformational opportunities across an entire healthcare field and is bound to generate an even more impressive impact in the foreseeable future. Nevertheless, it remains fraught with numerous drawbacks and challenges. Therefore, this study aims to determine the current global concerns and future directions of ChatGPT in medicine to inform subsequent research and policymaking.
Methods:
This study retrospectively analyzed the global attention and development patterns of ChatGPT in different medical disciplines and geographical regions. Furthermore, based on machine learning algorithms, it revealed the current global concerns and future directions of ChatGPT in the medical field.
Results:
ChatGPT enjoyed a favorable development trend (Growth Rate Per Month: 26.97%) and global cooperation (International Co-authorships: 25.09%) in medicine. Internal Medicine was the best developed, while Surgery [odds ratio (OR), 0.761; 95% confidence interval (CI), 0.608-0.954; P = 0.018], Health Care (OR, 0.744; 95% CI, 0.583-0.950; P = 0.018), Medical Informatics (OR, 0.622; 95% CI, 0.433-0.893; P = 0.010), Radiology (OR, 0.625; 95% CI, 0.433-0.901; P = 0.012), Public Health (OR, 0.611; 95% CI, 0.416-0.896; P = 0.012), and Oncology (OR, 0.571; 95% CI, 0.346-0.943; P = 0.029) needed further development. The unsupervised hierarchical clustering algorithm divided the global concerns of ChatGPT in medicine into six clusters, among which Cluster 2 (The Applications of ChatGPT in Oncology Patient Management and Decision-Making) is the emerging research cluster, and Cluster 4 (The Accuracy and Safety of ChatGPT in Health Information Recommendation) achieves maximal impact. With the Walktrap algorithm, we found that ethics [relevance percentage (RP) = 82.1%, development percentage (DP) = 92.9%] is well developed but still leaves numerous pending issues, and medical education (RP = 100%, DP = 32.1%) and clinical decision support (RP = 89.3%, DP = 35.7%) are highly relevant but under-developed with ChatGPT, highlighting their impressive future research prospects. More importantly, through a comprehensive analysis of ChatGPT's attention, application, and impact in different regions, we found that underdeveloped and resource-poor regions have little, which will exacerbate global health inequalities, emphasizing the urgency of the relevant policy formulation and international assistance.
Conclusions:
This study revealed ChatGPT's global attention and development patterns in different medical disciplines and geographical regions, its current global concerns, and future directions. This information will provide a critical reference for subsequent research and policymaking on LLMs throughout the entire field of medicine.
